Draft "Unsafe fp optimizations" project description.

[dewar@gnat.com]

>>Hence the chance is more of the order of 1/MAX instead of 0.5.

I don't buy that reasoning I must say :-)